DIGITAL
INTERFACES FOR URBAN SPACES
Master’s Thesis written by Lasse Hulgaard, 77284 Frederik Moesgaard, 42601
Supervised by Mads Bødker
Associate Professor Dept. of Digitilization
November 2020 MSc in Business
Administration and E-business
Copenhagen Business School
Master’s thesis Authors Lasse Hulgaard lahu12ab@student.cbs.dk Student#: 77284 Frederik Moesgaard frmo11ab@student.cbs.dk Student#: 42601
Business Administration and E-Business Copenhagen Business School
Denmark Title
Digital Interfaces for Urban Spaces Data submitted
Nov 16th, 2020 Supervisor
Associate Professor Mads Bødker Department of Digitalization Copenhagen Business School Characters and pages 97.339 / 43
CONTENTS
PREFACE 3
INTRODUCTORY CHAPTER 7
1 INTRODUCTION 8
2 THEORY 9
2.1 Related Work 9
2.2 Studying Humans and Technology ‘In the Wild’ 10
2.3 Incidentally Co-present Persons 11
2.4 Living with Robots 11
2.5 The Turn to Social Sciences 12
2.6 How Humans Make Sense of Robots 14
3 EXPERIMENT 15
3.1 Physical Environment 16
3.2 Mobile Robot Platform 16
3.3 Wizard of Oz Simulation 16
3.4 Breaching Experiments 17
3.5 Participants and Data Collection 17
3.6 Method of Analysis 17
3.7 Ethical Considerations 18
4 RESULTS 18
4.1 Types of InCoPs 18
4.2 Robot Membership Categories 19
4.3 How InCoPs Experience Robots in Natural Environments 20
5 DISCUSSION 25
5.1 Methodological Findings 26
5.2 Limitations in Methodology and Future Work 26
5.3 Contribution to HRI 27
6CONCLUDING REMARKS 28
APPENDICES A.1 Field Notes A.2 Field Footage
A.3 Documentation of Analytical Process A.4 Interview Guide
A.5 Paper 2: Involving Users in Sound Design
PREFACE
With an aim to make a contribution to the field of Human-Computer Interaction (HCI) and Human-Robot Interaction (HRI), our thesis format is an academic paper conforming with current ACM standards in terms of length, structure, and format. We intend to submit the paper to one of the following journals or conferences:
IEEE Transactions on Affective Computing, ACM Conference on Designing Interactive Systems (DIS), ACM Conference on Human Factors in Computing Systems (CHI), ACM Conference on Designing for User Experiences (DUX), ACM Conference on Ubiquitous Computing (UbiComp), HCI International, ACM/IEEE International Conference on Human-Robot Interaction.
The audience of the paper is intended to be researchers and practitioners in the field of HCI and HRI.
Structure of Thesis
The thesis is organized in three parts: First, an introductory chapter will situate our paper in a broader context. Our aim with this chapter is to outline our motivational background and reflections on how this paper fits in our academic context. As our focus has been on IT research and systems design, we will also use the introductory chapter to argue how the thesis can be seen from a business and policy perspective. Second, we will present our paper, Incidental Encounters with Robots. Finally, we will present our appendices including our research documentation. A second paper, Involving Users in Sound Design1, which we presented at HCI International 2020 conference, is included in the appendices.
This paper was co-authored with our supervisor, Associate Professor Mads Bødker. The reason for including our previous paper is to highlight that both papers are part of a bigger research agenda investigating how to shape future digital interfaces for urban environments. Both papers have been concerned with how such interfaces can be understood and developed from a human-centred perspective. Our current paper is a continuation of some of the processes that we started to explore in our previous paper.
Before moving on, we will present the abstracts of our two papers to set the scene for what to expect when reading our thesis.
Abstracts of Papers
Paper 1: Incidental Encounters with Robots.
Mobile robots will soon be part of our urban public environments, yet little is known about how humans and robots interact in such unpredictable and complex social settings. This paper reports on a qualitative field study of more than three hundred ‘incidental’ human-robot encounters in a public outdoor space. We have conducted an ethnomethodologically informed ethnographic inquiry using breaching experiments and membership categorisation analysis to reveal how Incidentally Co-present Persons (InCoPs) interact with and make sense of robots in an urban environment. Through our analysis of observations and interviews, we could identify six different types of InCoPs.
Additionally, we have identified seven distinct categories, which people used to make sense of the robot. Each of these categorisations led people to adopt certain expectations and behaviour during the interaction, which has been reported on. Our results further show how ethnomethodology can provide a new perspective to study the behaviour and experience of people incidentally meeting a robot in a public space. With this paper we propose a new approach in Human-Robot Interaction research, where robots are seen from the perspective of being designed to ‘co-exist’ with people and integrate naturally into the urban environment.
Paper 2: Involving Users in Sound Design. (Appendix 5)
Sound plays an important role in our well-being, our experience of the world around us and our understanding of products, services and interactions. Sound affects our sense of place, and it can modulate our feelings, agency and
1 https://doi.org/10.1007/978-3-030-49713-2_28
attention. In a world of increasingly ubiquitous digital technologies, sound may prove a valuable resource for sense making as well as experience- and UX design. Yet the possibilities and challenges of user participation in sound design processes are not well understood. This paper reports on a pilot study examining how participants can be involved in different phases of a sound design process. The results and reflections aim to help researchers and designers in an effort to better understand some of the dynamics of moving from a largely expert driven approach to sound design towards a more user-oriented and participatory approach.
INTRODUCTORY CHAPTER
Following the two abstracts presented above, we will now explain how our thesis can be seen from a broader technological, societal, and human perspective. We will then briefly elaborate on our overall research approach and methodological considerations. Finally, we will share our reflections and provide a future outlook.
Technology in Urban Environments
Strolling through an urban environment today means being surrounded by technology, machines and interfaces. Doors, lights, cameras, and speakers come to life as you approach them, often automatically and without you even noticing their existence. We are currently in the early days of building intelligent urban environments, and within the near future, digital and ubiquitous technologies will be an increasingly integrated part of our public spaces. As you commute with the underground metro, sonic interfaces embedded in the walls might inform you about the current weather conditions above. And as you step outside, you might then be greeted by a robot collecting cigarette butts from the street or delivering food to a family around the corner. At the periphery of your attention these technologies may inform you, but they may also influence what you do, how you do it, or how you feel.
As aspiring designers, we are concerned with how these future urban technologies will become woven into people’s complex lives, how people make sense of them, and how they will coexist with the abundance of other technologies that we surround ourselves with.
Ubiquitous Computing and Intangible Interfaces
In the year we were both born, 1991, Mark Weiser presented his vision for what he called ‘Ubiquitous Computing’. In his paper The Computer for the 21st Century, he imagined how we would one day be surrounded by technologies that
“weave themselves into the fabric of everyday life until they are indistinguishable from it.” [1, p. 94]. A few years later, in 1995, and together with his Xerox PARC colleague, John Seely Brown, he published the paper, The Coming Age of Calm Technology, in which they predicted that a scarce resource of the 21st century would not be computing power, but our attention. Building on top of the work by PARC anthropologists, such as Lucy Suchman, their view of technology was from a human-centred perspective. Predicting that the number of devices would increase, their interest was in how technology could be designed to sit at the periphery of a person’s attention. This would allow it to stay out of focus and come to the stage when needed, avoiding constant interruptions. They called this calm technology.
If we fast forward to today, just as Weiser and Brown predicted, we are surrounded by digital devices fighting for our attention. Not only in the home or at work, but also in public spaces. As advanced technologies are packaged into ever smaller devices, more everyday objects have become connected, intelligent, and autonomous. As a designer, the challenge is how to design these future interfaces. As technologies “weave themselves into the fabric of everyday life”, they become surface-free or ‘faceless’ as Janlert & Stolterman [2] call it. As a result, such interfaces are vastly different from the ones we are used to designing and using, and they bring with them new questions about complexity, control, and interaction. The challenge is that these interfaces are often more abstract and intangible. They are not bound to screens and buttons, but it may be thought of as a ‘force field’. This is described further by Janlert & Stolterman [2]:
“Faceless interaction in general makes way for interaction that is not restricted by the location or direction of a well- defined, well-delimited source or target. (...) “users” are immersed in an ecology in which they are not conceived as interacting with particular, targeted objects one at a time but rather are moving in situational and interactional “force fields,” causing
minor or major perturbations in their environment by their moves and actions while being guided, buffeted, seduced, or affected in any which way by constant movements and changes in their environment taken as a whole.” (p. 532).
When people wander in and out of such force fields in urban environments, they interact both directly and indirectly with a variety of technologies, such as robots, screens, sensors or speakers. It is their experience of these interactions that we have been interested in investigating through this thesis. In neither of our two papers, Incidental Encounters with Robots and Involving Users in Sound Design, has our aim been to come up with a finalised interface design. Instead, we have been concerned with developing new ways of studying people’s experiences with intangible interfaces (robots and speakers, respectively) from a human-centred perspective.
It is our position that designing more humane interfaces is more pressing than ever before. We further argue that this is particularly important when designing technologies for urban environments, where people will not necessarily have chosen to ‘use’ the technology and may therefore not have any interest in interacting with it at all. In the case of the latter, these people are simply co-existing with the technology.
Our approach to studying intangible interfaces
Our work has been underpinned by the sociotechnical ontological stance that our understanding of the surrounding world is a product of collectively held imaginaries that come to life from peoples lived experiences with technology. We may construct knowledge individually through interpretation of lived experiences, but our sense of reality is constrained by social conventions, as well as what we believe our surroundings could be or do. As such, our experiences of our surroundings and technology are somewhat controlled yet varied. This position highlights the need to actively engage users and find methods that uncover how people experience technology.
However, involving users in the design or research of future technologies is challenging, due to the lack of real-world use and practice. So, in our quest to study people’s experiences of interacting with ‘future’ technologies in urban environments, we have deployed early prototypes of the technology in order to learn from concrete experiences. This has been a way to do a contextual inquiry by ‘faking’ a real-world experience with the technology.
In order to make the ‘fake’ experience with the technology as real as possible, it has been important for us to deploy these prototypes ‘in the wild’, i.e. in a setting, which is as close to a real world setting as possible. Additionally, in both papers, we have recruited participants who knew nothing or only very little about the experiment beforehand.
In our previous paper, Involving Users in Sound Design, we used variations of prototyping and ethnographic field methods, such as sound walking, sound sketching, and sound prototyping, to inform the design of sonic interfaces for the Copenhagen Metro. This project revealed to us the value of ‘in the wild’ studies and probing techniques in combination with an ethnographic inquiry to inform the design of an intangible interface (sound). Moreover, we found that people had very different experiences despite being exposed to the same technology. Writing this previous paper motivated us to further explore the dynamics of interacting with technology in urban environments and has now led us to investigate the interactions between humans and robots in public spaces. Here, we have taken a similar approach and deployed a mobile robot in a public space and treated it as breaching experiments to provoke or ‘call forth’ people’s expectations to the technology, as suggested by Crabtree [3].
So, why robots?
As an urban technology, we believe robots will have a more substantial impact on human experience compared to most other ‘intelligent’ technology deployed in public spaces, which are often more subtle in their appearance. Deployment of new robots in public environments grew by 44% globally in 2019, serving purposes such as inspection, maintenance, defence, logistics and agriculture [4]. Meanwhile, reports from Europe and the US suggest that the broader public is not yet willing to accept robots for everyday use [5, p. 201]. Similar studies have shown that people report high levels of robot anxiety and other negative attitudes towards robots [5]. Little is known about the underlying dynamics
constituting these experiences with robots, but these studies suggest that introducing robots on a larger scale might not be entirely problem-free.
Despite the recent and continued advances in technologies such as artificial intelligence (AI), machine learning, and 5G connectivity, roboticists still face a challenge of incorporating the common-sensical human behaviours and expectations into robots. As a result, robots find it hard to navigate in social environments, as these are often unpredictable and sometimes guided by ‘unwritten’ or ‘intuitive’ social rules and assumptions. The automatic and unconscious actions performed by humans in social environments are much more complex and harder to reverse- engineer than logical reasoning. A phenomenon also known as Moravec’s Paradox. Generally speaking, tasks that are completely effortless for humans are often hard for computers, whereas tasks that are difficult for humans are usually easy for computers. As such, robots may be ready to replace humans in a variety of practical tasks, but they are not yet capable of interacting seamlessly and naturally when encountering the unpredictable.
Hence, our intent with our current paper, Incidental Encounters with Robots, is to contribute to the field of HRI by exploring some of the dynamics in play when a person incidentally meets a robot in a public space. By doing so, we aim to include a sensibility to the role of context and social setting in human-robot encounters, which we believe is lacking in existing HRI literature.
Master of Business Administration and E-Business
As master’s students of Business Administration and E-business, our approach to the development of technology and IT systems is not only about how things can be achieved faster and cheaper, but also about exploring potential avenues of future technologies. During the course of the programme, we have been working with the development of future digital technologies from both a technological, business, and policy perspective. Our thesis has been concerned mainly with the (socio-)technological perspective. In this section, we will briefly argue why this approach to understanding robots fits into a broader context including business and policy.
Robots from a policy perspective.
Use of autonomously guided robots are currently prohibited in public environments, not only in Denmark, but in most parts of the world. Many robots are, from a purely technical and functional perspective, ready to be implemented on a larger scale, but one of the greatest barriers remains the regulation and legislation. Legislators must address a multitude of practical and ethical questions and make sure that benefits are weighed against potential danger or negative effects.
For regulatory efforts to be efficient, it is critical that they are based on research from real world practice, rather than assumptions and untested hypotheses. The result may be over or under regulation leading to either lack of benefits realisation or potentially hazardous outcomes. Finding the appropriate balance in regulation is made easier with a solid base of research and practice to back up the law-making. The legal challenges of releasing robots into public spaces to interact with the ‘real world’ are comprehensive and must be addressed from a multidisciplinary perspective. Some legal questions, such as: ‘Who bears the responsibility if a robot does damage to a person or property - the robot manufacturer, robot owner, robot operator, or someone else?’, will not be possible to answer meaningfully through studies like ours.
However, we believe that questions, such as: ‘What should the speed limit of robots be and should it change depending on the context?’, ‘In what public spaces should robots be allowed?’, or ‘In what social contexts should robots be prohibited (e.g. large gatherings of people)?’ could benefit from qualitative real-world experiments like ours.
Robots from a business perspective.
The commercial potential of robots is vast. While industrial robots have been used since the 1930s, service robots are only recently starting to gain serious traction. According to a 2019 report from Deloitte, the market for service robots is growing much faster than that for industrial robots. They predict that service robots will pass industrial robots in terms of units in 2020 and revenue in 2021 [6]. However, just as organizations care about the capabilities, behaviour and attitude of their employees, so should they when hiring robots. Acknowledging the importance of how humans and
robots interact, studies like ours may provide useful insights to answer questions, such as: ‘When should we use robots and when not?’, ‘What can we expect, when employing robots?’, ‘What type of robot should we use?’, or ‘What kinds of tasks are meaningful for a robot to perform?’. These questions are critical in order to understand the commercial potential in a given practical context and should, in our view, predominantly be informed through concrete practice. As Matarić [7] points out: “the majority of HRI research is entirely disconnected from, or even aware of, commercial and societal impacts.” (pp. 1).
To summarise, we argue that a more nuanced understanding of the interaction between humans and robots is critical, not just from a design and technology perspective, but also from a policy and business perspective. The main reason being that it enables us to test the assumptions about what we think might happen, when robots are released into the real world.
Our reflections and a brief look into the future
The reason for choosing the ethnographic field ‘urban spaces’ rests on our interest in design, architecture and urban development on one hand, and technology on the other hand. What we have come to realise during our two-year master’s degree is this: Technology is increasingly moving from our workspaces into public environments, but while this shift is happening, we have limited understanding of how these technologies impact our society and personal lives.
Technology is often developed from a perspective of efficiency and automation, but we are more interested in what else technology could and should be, and how they integrate naturally in our social environments. In writing this thesis, we have realised the importance of understanding the human and social context in detail, when designing and developing new technologies. As we both share a passion for design and technology, it is our hope (and belief) that this thesis has equipped us to contribute - either academically (PhD) or professionally - to the shaping of future technologies for urban environments.
Incidental Encounters with Robots
LASSE HULGAARD Copenhagen Business School FREDERIK MOESGAARD Copenhagen Business School
Mobile robots will soon be part of our urban public environments, yet little is known about how humans and robots interact in such unpredictable and complex social settings. This paper reports on a qualitative field study of more than three hundred ‘incidental’
human-robot encounters in a public outdoor space. We have conducted an ethnomethodologically informed ethnographic inquiry using breaching experiments and membership categorisation analysis to reveal how Incidentally Co-present Persons (InCoPs) interact with and make sense of robots in an urban environment. Through our analysis of observations and interviews, we could identify six different types of InCoPs. Additionally, we have identified seven distinct categories, which people used to make sense of the robot. Each of these categorisations led people to adopt certain expectations and behaviour during the interaction, which has been reported on. Our results further show how ethnomethodology can provide a new perspective to study the behaviour and experience of people incidentally meeting a robot in a public space. With this paper we propose a new approach in Human-Robot Interaction research, where robots are seen from the perspective of being designed to ‘co-exist’ with people and integrate naturally into the urban environment.
CCS CONCEPTS • Human-centered computing • Human computer interaction (HCI) • Empirical studies in HCI Additional Keywords and Phrases: Human-robot interaction, Ubiquitous computing, Ethnomethodology, Breaching experiments, Membership categorisation analysis, InCoPs, In the wild
1 INTRODUCTION
Advancements in robotics and artificial intelligence have allowed for an increased presence of robots in our everyday lives. Up until now, these machines have been limited to work within the confines of our home, workplace or the facilities of the robot operator. Soon, however, it may no longer be unlikely to accidentally meet a stranger in the park in the form of a robot. Now pause for a second and think about what that experience might be like. Taking departure in the social sciences, we seek to explore people’s experience with robots in everyday social environments. We argue that a better understanding of these experiences is needed to inform design, development and regulation of robots in public spaces. What we offer is a new approach to the field of Human-Robot Interaction (HRI) by moving beyond the notion of being a ‘user of a robot’, towards the more open and relational experience of ‘co-existing with robots’.
The HRI community has published a vast amount of literature to support the design and development of robots, but two main limitations to current literature persist. Firstly, most HRI research reports on the interactions between robots and primary users of robots. In a real-world setting, however, the majority of the robot encounters will be with people who are incidentally co-present in the situation. Rosenthal-von der Pütten and colleagues [8] call this interaction the
“forgotten in HRI” and recently coined the term Incidentally Co-present Persons, or simply InCoPs, to describe this group of ‘passive’ bystanders. Secondly, most HRI studies are performed in controlled laboratory settings, which fail to emulate the complexity of the real world. Similar to what has been suggested by several HRI and HCI2 researchers, we believe such studies reflect an oversimplified view of HRI (e.g. [7, 12]) We believe the time is ripe to ask questions, such as:
What happens when a robot navigates through a crowded public space occupied by not only the ‘average’ user, but also
2 Human-Computer Interaction
dogs, cars, cyclists, scooters, wheelchair users, abusive children, or perhaps a fellow robot? How do these encounters unfold, how do people make sense of them, and how do we make sure to include them in our design considerations?
With this paper, we seek to improve the understanding of human-robot interactions, by uncovering and elaborating on the experience of meeting an unknown robot in a public space. Our intention is not to put forth a bulleted list of implications for design, which is often the focus in both HCI and HRI research. Instead, and similar to what has been suggested by [13], our focus has been on the insights revealed from an ethnographic inquiry. We aim to equip designers with a better sense of the social dynamics at stake, allowing for more human-centred design of robots.
Taking the stance that an experience is a product of an interaction and a process of ‘making sense’ within the human mind, we will proceed and answer the following research question:
§ How do InCoPs interact with and make sense of mobile robots in a public outdoor space?
By ‘interact’ we refer to the exchange of actions between people or objects, and by ‘make sense of’ we refer to a “a motivated, continuous effort to understand connections (which can be among people, places, and events) in order to anticipate their trajectories and act effectively” [14, p. 71].
In our approach to answer this question, we have turned to the ethnographic methods of observations and interviews.
Additionally, to enrich our insights, we have turned to ethnomethodology, and specifically breaching experiments and membership categorisation analysis.
Answering this question will not only illuminate how InCoPs interact with and make sense of robots, but also what InCoPs expect from robots as they encounter them. Additionally, we believe that our attempt to inform the research agenda may provide a methodological contribution by answering how we can study interaction between InCoPs and robots in the real world. We argue that investigations and examinations of robots and incidentally co-present persons have never been more needed than now. We believe such investigations enrich our field and might be inspiring to designers in their quest to shape future technologies.
The paper sets out by providing an overview of related work and elaborate on the need to study InCoPs in natural environments. Next, we argue why ethnography and ethnomethodology are well suited to inform our research question.
This section forms the theoretical underpinnings of our study. We will then turn to the practicalities of our field experiment, and finally present our results of this study. We round off by discussing our findings and present our concluding remarks.
2 THEORY 2.1 Related Work
When seeking to understand how humans experience robots in the wild, a natural first step is often to turn to existing literature. The field of HRI can be described as the study of humans, robots and the way they influence each other [15].
The human-centred branch of HRI has typically been interested in the psychological effects of specific robot attributes.
Such robot attributes include effects of robot speed (e.g. [16, 17]), robot navigation (e.g. [18]), and sound (e.g. [19, 20]).
The psychological effects studied include users’ level of trust (e.g. [21, 22]), comfort (e.g. [23, 24]) or safety (e.g. [25, 26]).
This body of literature leaves no doubt that robot behaviour and overall design have a significant impact on the way robots are perceived by humans. What we do suggest, however, is that robot behaviour and design can only partly explain the overall experience of interacting with a robot. As a result, we see several limitations in the transferability of these studies to the context of interactions between humans and robots in natural social environments.
Firstly, most studies are performed in a laboratory setting, which often fail to anticipate and simulate the complexity of the real world. Findings from laboratory environments have also been found to differ markedly from field experiments,
where participants are behaving of their own volition [27]. Thus, results from a laboratory experiment could likely lead to a proposed robot behaviour that works only in the lab but not in a real-world deployment. Jung & Hinds [9] have noted the lack of ‘in the wild’ research and argue that at this point, it is not possible to predict how humans and robots influence each other over a variety of contexts, as there are simply too few studies to provide generalisable theories within the field.
Secondly, the interaction in focus has mainly been between robots and primary users, rather than InCoPs. This has been a main concern for Rosenthal-von der Pütten and colleagues [8], who highlight the lack of studies investigating the somewhat unpredictable and varying behaviour from different groups of InCoPs. Additionally, the quantitative and confirmatory approaches used in many HRI studies, do little to inform about the experiential aspects of human-robot encounters.
With this paper, we aim to contribute to the field by addressing these current limitations. In the following section, we will briefly clarify the dynamics of ‘in the wild’ studies in HCI and HRI, the challenging nature of investigating InCoPs, and our reason for turning to social sciences to address the two.
2.2 Studying Humans and Technology ‘In the Wild’
The idea of studying technology use in ‘natural’, ‘situated’, and ‘in the wild’ settings, as distinct to lab-based studies, is now starting to receive more attention in HCI and HRI communities [28], as it enables researchers to pursue the more complex but messier questions around humans and technology.
Studying technology use in a real-world setting goes back decades. Early works in this field were shaped by anthropology and ethnography, with researchers in the early 1980s attempting to introduce a sensibility to the role of context when developing technologies. Work by Xerox PARX anthropologists, such as Lucy Suchman’s well known distinction between plans and situated actions in HCI [29]3, played a significant role in bringing this new perspective to how technology could be understood from a more human and real-world perspective. As Rogers and colleagues [30]
write:
“while lab approaches can highlight variables of interest, they may miss important factors that only come to light in in- the-wild settings” (p. 370).
Lab-based studies may be good at sensing and describing specific technology attributes or certain aspects of human behaviour, they are poor at capturing context of use.
In the field of HCI, lab-based versus ‘in the wild’ studies have been discussed extensively, especially in the past 15 years. As most HCI and HRI research has been conducted and evaluated in controlled lab-based settings (see [7, 11, 31, 32]) we find it necessary to briefly comment on the background and dynamics of both lab-based and ‘in the wild’ studies.
Specifically for usability evaluation, Kjeldskov and colleagues [31] found lab-based studies to be superior to ‘in the wild’ studies for a number of reasons. Their comparative study revealed that the lab-based setting was generally better at identifying usability problems. They further argued:
“lack of control in field-based evaluations makes it challenging for evaluators to conduct field evaluations in practice and to make sure that every aspect of the system is covered” (pp. 11).
In their revisit of the paper published 10 years later [33], they extend this discussion to cover broader aspects of HCI, and suggest that:
3 Other notable contributors include Jeanette Blomberg, Brigitte Jordan and Julian Orr.
“Since no answer to the lab versus field question seems to be found, we have argued that the important question is not if or why one should do lab or field studies, but rather when we should do what, and how we should then do it. As input to moving the discussion of empirical methodology forward, [...] we should embrace field studies that are truly wild and longitudinal in nature in order to fully experience and explore real world use” (pp. 50).
In relation to HRI research specifically, Matarić [7] explains the overrepresentation of lab-based studies by the difficulty of getting real-world HRI studies published. He presents three major reasons for this. First, ‘in the wild’
problems involving real people can be messy and do not always lend themselves to clear formalisation. Second, ‘in the wild’ studies are costly and time consuming (which delays publication), because they often require long-term commitment to interacting with participants outside of the lab. Third, ‘in the wild’ studies are difficult to fit into expectations for ‘clean’ experiment design with variables that are easy to isolate and measure. These can all be reasons why real-world studies do not immediately lend themselves to top publication venues.
The importance of context, however, is illustrated in the study by Brignull & Rogers [34], in which they found that people in public spaces were hesitant to interact with unfamiliar technology (in this case large interactive screens) because people felt self-conscious. Their results indicate that people felt uncomfortable knowing that their actions and their effects were highly visible to strangers, which affected their willingness to interact and use the technology [30], [34]. In another example, Rogers and colleagues [32] found that changes in the physical environment, such as the time of year or weather conditions, can have a significant impact on how the interaction is experienced.
With this paper we seek to report on people’s experiences of incidentally meeting a robot in a public space and argue that an ‘in the wild’ approach is best suited to address this research agenda. Rather than only attending to the design and behaviour of the robot, we are also interested in the contextual facets that may shape the interaction.
2.3 Incidentally Co-present Persons
We share the position of Rosenthal-von der Pütten and colleagues [8] that Incidentally Co-present Persons, or InCoPs, are often overlooked in existing HRI research. This is important to note because as robots are increasingly released into public spaces, spontaneous encounters will in many cases make up the majority of a robot’s encounters. To this date, only a small volume of research has been dedicated to understanding the InCoP’s perspective in HRI (e.g. [21, 35–37]).
This paper aims to join this conversation.
One of the more significant differences between InCoPs and primary users is that of expectations. InCoPs rarely expect the interaction to take place. They do not necessarily know the purpose of the robot and may not be interested in interacting with it at all. The attention of InCoPs might initially be elsewhere; they may be engaged in conversation, using their mobile phone, trying to find their way, or preoccupied with other activities. In other words, interacting with a robot is not their top priority. However, InCoPs often initially engage in an unfocused interaction. This has been described by Erwin Goffman [38], whose works have been foundational in the understanding of interactions at a micro- sociological level. Goffman argued that the behaviour of bystanders is constrained by etiquette and can be associated with the phenomenon of ‘civil inattention’ when in social situations [39]. As a result, their behaviour is influenced by their role in the social setting.
As a bystander, or what we refer to as an InCoP, the interaction with a robot will usually be initiated in the periphery of their attention and may (or may not) later proceed to the centre of their attention, going from an unfocused to focused interaction. How this interaction unfolds, what people expect of the interaction, and how the interaction should be designed naturally become important questions for HRI research to address.
2.4 Living with Robots
As robots are woven into everyday life, it is important that we explore the different roles they are to play. Similar to what has been suggested by Reeves & Nass [40], we consider robots as social actors. This has been our point of departure on our journey to understand the interactions between robots and InCoPs. Reeves & Nass [40] write:
“Computers, in the way that they communicate, instruct, and take turns interacting, are close enough to human that they encourage social responses” (p. 22).
Based on their politeness studies, which show how people have a tendency to be polite towards technology, they argue that “(...) the biggest reason for making machines that are polite to people is that people are polite to machines. Everyone expects reciprocity, and everyone will be disappointed if it’s absent (...). It’s not just a matter of being nice; It’s a matter of social survival.” (p. 28-29).
If we accept the condition that technology should behave politely, i.e. behave in a respectful and considerate manner, the question then becomes: How does a robot behave politely towards InCoPs? Should it stay in the periphery of your attention and ‘disappear’ into the background, as suggested by Mark Weiser [1] in his vision for ubiquitous computing?
Should it ‘stand out’ and engage in playful interactions with people (see e.g. [41, 42])? Should the aim be to design robots with a sense of ambient intelligence (see e.g. [36, 43])? Should the interaction allow for multiple meanings and interpretations (see e.g. [44, 45]), or should the aim perhaps be to create a moment of meaning and pleasure (see e.g. [46, 47])?
These possible scenarios all represent the varying facets of interactions between humans and technology.
Considering these scenarios, how a robot is designed should not only take into account the function of the technology and its specific context, but also who it will interact with. This is when designing technologies for complex social environments, where the possible types of encounters are endless and unpredictable, becomes incredibly challenging.
Nevertheless, such considerations deserve special attention when designing technologies, which need to naturally co- exist in the same environment as people who just ‘happen to be there’.
Our final remark in relation to the special nature of InCoPs is that of Janlert & Stolterman’s [2] notion of the ‘faceless’
interaction. We argue that the interaction between InCoPs and robots can be seen as an intangible field, which is affected by the entities within it, rather than a traditional one-to-one interaction. Humans and robots may co-exist in a dynamic ecology, where moves of one object affect the entire environment or ‘field’, without holding a specific target as a recipient. Thus, we need to move beyond the notion of humans in HRI as ‘users’ of a technology, towards the idea of people ‘being-with’ or ‘co-existing’ with technology.
To understand how robots should present themselves in this ecology, we must find ways to explore and articulate this intangible interaction between humans and robots. This realisation has motivated a shift from the lab into the wild, but also encouraged HRI researchers to adopt new human-centred methodologies from fields such as ethnography and ethnomethodology.
2.5 The Turn to Social Sciences
As previously mentioned, ‘in the wild’ studies have led to an increased interest in social science methodologies as a resource to guide design, in particular ethnographic inquiries. In this section we elaborate why this movement towards the social sciences is needed to understand how humans interact with and make sense of robots.
2.5.1 Ethnography.
Ethnographic research is a qualitative method used to observe and interact with users in their own environment. It is the branch of social science, and specifically anthropology, which involves trying to understand how people live their lives and make sense of their experiences. In HCI and the design of IT systems, the role of the ethnographic researchers is that of a translator between the group being studied and the designer [48]. In this translation process, the ethnographer makes motivated choices about what to study, who to observe, and what activities to record. As such, ethnography can be used to understand how people make sense of experiences through the interpretation of accounts from fieldwork [47]. In relation to systems design, ethnographic methods hold the promise of uncovering the social practices in which technology is embedded, and by doing so, it allows us to step inside the situation and describe it, as it unfolds [49]. This process often involves field observations and user interviews, followed by an analytical approach to grouping and coding
the raw data from the field. By abductively interpreting and generalizing accounts, it is possible to derive actionable insights and ideas to directly inform design [50].
2.5.2 Ethnomethodology.
Another, but less common approach is found in the field of ethnomethodology, pioneered by Harold Garfinkel. Relying on detailed ethnographic field observations, such as video recordings, the ethnomethodological approach takes a deeper dive into the details of the interaction. It attends to the practices, which humans use to perform taken-for-granted actions at a micro-sociological level [51, 49, p. 6, 52]. As such, ethnomethodology provides a way to unfold how people ‘make sense of’ their experiences. Ethnomethodology rests on the idea that humans possess a capacity to perform everyday behaviour in accordance with a shared set of expectations, and thereby produce a state of social order [53, 54, p. 150].
Sacks argued that we use a practice of ‘membership categorisation’ [55]. By assigning people, events and objects into membership categories and drawing on our shared tacit knowledge of such categories, we form expectations and know how to behave in everyday interactions for social order to prevail. Though not explicitly stated in ethnomethodology research, we refer to this as a process of ‘making sense’4. We adopt the underlying premise that the process of ‘making sense’ is a continuous effort as all interactions are context-dependent and context-renewing [51]. However, as this process is an ingrained and implicit process in the human mind, it is not easily examinable. For this reason, practitioners of ethnomethodology rely on breaching experiments.
Breaching experiments, typically associated with Garfinkel, are experiments that seek to examine people’s reactions to violations of commonly accepted social norms. They are not positivist experiments that try to prove a cause-effect relationship but rather a tool to ‘call forth’ everyday practices and expectations and thereby make them available for analysis [56]. In a classic example, Garfinkel asked his students to act like guests in their own home, resulting in bewilderment and confusion among their family members, expressed both verbally and non-verbally [51, p. 47]. He argued that when the collectively held convention of behaviour, in this case between family members, is breached, their reactions become testimonials of their hidden expectations.
In several HCI and HRI studies leveraging breaching experiments, researchers found that deploying technology ‘in the wild’ can be a breaching experiment in itself if the technology is unfamiliar [3, 57]. However, it is up to the researcher to define the appropriate breaching experiment, depending on the context and technology. An example of the use of breaching experiments in HCI is described by Crabtree [3]:
“The ethnomethodological notion of breaching experiments has recently been employed by Steve Mann (2003) in his remarkable exploration of computer wearables and surveillance technologies. Mann employs breaching experiments to actively create situations of uncertainty, bewilderment, anxiety and confusion in order to bring into question everyday structures of surveillance, governance, and control.” (pp. 60).
Similarly, our intent is to employ breaching experiments to bring into question the everyday structures of robots in public spaces.
Breaching experiments are normally used in combination with conversational analysis and membership categorisation analysis [53, p. 663, 55]. These modes of analysis look at every miniscule detail in interactions to describe how we make sense of things by using our pre-existing and continuously updated knowledge [55]. In membership categorisation analysis, a special attention is paid to how we assign people, objects and events into categories and what expectations and actions these categories produce. Ethnomethodologists will argue that these raw descriptions lead to rich insights that are truly human-centered compared to what can be achieved through more interpretive ethnographic strategies, as these will be secondary accounts reflecting the interpretations of the researcher [49, p. 15, 52].
4 Not to be confused with the terms ‘sense-making’ by Dervin and ‘sensemaking’ by Weick.
However, using insights from conversation analysis or membership categorisation analysis to inform design has proved difficult. The strict methodology of such analyses disregards any attempts to theorize and group data [58]. It simply presents raw field accounts, which requires the reader to be well-versed in ethnomethodological descriptions [52]. As a discipline, ethnomethodology does not really aid in shaping the future, which makes it somewhat unconstructive [49, p. 6, 52]. In turn, it has served more as a critique of systems and their shortcomings [56].
2.5.3 From Design Critique to Design Practice.
The challenge of deriving rich insights from ethnomethodology, while still allowing a more interpretive approach to data analysis has driven researchers towards the notion of ‘ethnomethodologically informed ethnography’ (e.g. [49, 59]).
In such approaches, researchers do not follow the strict guidelines and limitations of traditional ethnomethodology, but rather use elements of ethnomethodology in their ethnographic inquiry. This has been found to be a more constructive approach, when the aim has been to inform design practice. The increased use of ethnomethodology in information technology (IT) research has led Button & Dourish [52] to develop a hybrid discipline, technomethodology. In their paper from 1996, Technomethodology: Paradoxes and Possibilities, they present and discuss three different approaches in which ethnomethodology can be used in IT research:
§ Learning from ethnomethodologists. In this approach, an ethnomethodologist is actively involved in the design process working closely together with the designer or design team. The rich insights are derived directly from the ethnomethodologist, as the designer bounces ideas and possible design directions.
§ Learning from ethnomethodological accounts. The ethnomethodologist produces and records ethnomethodological accounts, or written reports of analysis that can later be read and used by designers to inform their work. The ethnomethodologist makes no attempt to generalise or group data, which requires the designer to be well-versed in ethnomethodological writing.
§ Learning from ethnomethodology. In the third approach, the designer takes the role of the ethnomethodologist. Instead of the designer and ethnomethodologist being two separate entities coming from each their discipline, the two merge into one. Rather than have design and ethnomethodology ‘reach’
towards each other in the design process, their intention was to forge more foundational relationships between the two. This is what Button & Dourish refer to as technomethodology.
Since our aim with this paper has not been to develop a final design of a robot interface, we do not adopt the suggested hybrid discipline of technomethodology. Instead, our approach is more similar to what Button & Dourish [52] call learning from ethnomethodological accounts, with the notable difference that we have adopted a more interpretivist approach to analysing and grouping data, more similar to an ethnographic inquiry. It is our position that drawing on concepts from both ethnomethodology and ethnography is a constructive approach to inform our research agenda.
2.6 How Humans Make Sense of Robots
In the previous sections, we have elaborated on the importance of ‘in the wild’ studies and why there is a need to better understand how InCoPs interact with robots. Additionally, we have argued why a turn to the social sciences is necessary to inform our research agenda. We will now suggest how we can address our research agenda by proposing a framework summarising our theoretical foundation (Fig. 1). It is not prescribing a priori codes but serves as a lens to help researchers and designers investigate what kinds of knowledge people use when meeting a robot, what categories they assign the robot to and what this leads to in terms of expectations and behaviour.
Firstly, the model deploys the ethnomethodological perspective that any interaction results in a process of ‘making sense’ to reach social order. The model shows the reflexive and indexical nature of the process. Different sources of continuously updated knowledge are used to assign the robot to a membership category, which in turns forms expectations and behaviour.
To simplify things, we have divided the sources of knowledge into five categories. First, we adopt the distinction between local knowledge and working knowledge as used by Crabtree [3]. No clear definitions of these notions are presented in current literature, but we will define ‘local knowledge’ as the knowledge people have about the local context of the encounter, and ‘working knowledge’ as their pre-existing knowledge of (or previous experiences with) similar robots. We define ‘social role’ as the role people have in the specific situation and context of the interaction (e.g. being a parent or a police officer), as we believe this has an influence on how they categorise the robot. The reason for including
‘social role’ is informed by the works of Goffman [38] describing how people behave differently depending on the social setting. Finally, the continuously updated knowledge of robot behaviour is an essential source of knowledge, but also knowledge of design, which has been found important in lab studies has been included. We define ‘Robot Design’ as the physical appearance of the robot, and ‘Robot Behaviour’ refers to what the robot does and how it does it (e.g. speed or navigation).
Figure 1 below is a visualisation of this iterative process we suggest is emerging when InCoPs encounter and make sense of a robot.
Figure 1: Visualisation of how InCoPs ‘make sense’ of robot encounters.
Visualised by Hulgaard & Moesgaard (2020).
We acknowledge that the proposed model does not grasp the full complexity of how people make sense of robots, however we believe it serves as a useful tool to deconstruct the process of how InCoPs experience robots in natural environments. We will now turn to the practicalities of our research.
3 EXPERIMENT
We have conducted an ‘in the wild’ study in a public space in Copenhagen, Denmark. During this study we have relied on breaching experiments and Wizard of Oz simulations with an outdoor mobile robot. Data collection has been carried out by gathering ethnographic data in the form of notes from field observations and short in situ interviews with participants. Subsequently, this data was analysed using a variation of membership categorisation analysis and interpretative coding to reflect the reality of human-robot encounters.
You assign the robot to a category, or update category based on new input MEMBERSHIP CATEGORISATION New input from the robot, context,
or situation is received INPUT
The category influences your behaviour and expectations of the robot BEHAVIOUR AND EXPECTATIONS WORKING KNOWLEDGE
Your pre-existing knowledge about similar technology
Your knowledge about the local context and situation LOCAL KNOWLEDGE
Your social role in the context and situation SOCIAL ROLE
The physical appearance of the robot, i.e. how it is designed ROBOT DESIGN
The behaviour of the robot, i.e. what is does and how it does it ROBOT BEHAVIOUR
3.1 Physical Environment
The field study was carried out in an outdoor public space near the Copenhagen Business School campus in Denmark (Fig. 2) between the 28th September - 16th October 2020. The space is shared, characterised by being used by pedestrians as well as cyclists. The patterns of motion ranged from very determined (e.g. the businessperson on their way to a meeting) to wandering behaviour (e.g. the family out for an evening stroll). Students use the passage to walk between buildings on campus, and many cyclists use the passage to get from one bike path to the next. Hence, the space is a combined public and urban transit space offering a dynamic, complex and diverse experimental environment.
3.2 Mobile Robot Platform
A prototype of an autonomously guided robot developed by Capra Robotics (Aarhus, Denmark) was used in the field study (Fig. 2). The pneumatic wheels and patented chassis allow for driving in both indoor and outdoor conditions and across undulating terrain and rough surfaces. The model used, P1.3, is radio controlled and capable of driving at speeds up to 12 kmph. According to the robot manufacturer, the intended use cases for the robot include last mile delivery, humanitarian aid, inspection, security, and personal assistance. The robot in use was a ‘neutral’ looking robot, with no clear purpose, and with no cameras or sensors attached.
Figure 2: Top: The yellow line marks the boundaries of the location of the experiments.
Bottom left: Children interacting with robot. Bottom right: Closeup of robot.
3.3 Wizard of Oz Simulation
The breaching experiments were performed using the Wizard of Oz (WOZ) technique, a commonly employed technique in HRI experiments [60]. The WOZ is a prototyping approach that ‘fakes’ a working system (in this case an autonomously guided robot) in order to test relevant aspects of a system without spending unnecessary time or other resources. Using this technique allowed us to control the robot remotely and perform the desired breaching experiments, instead of programming the robot to perform them autonomously. A wireless speaker connected to a smartphone was placed inside the robot, which enabled the testing of sonic interactions. One researcher would control the robot while the other would secretly control the sound, simulating that the robot was able to make these sounds on its own. Figure 3 shows one of the experimental setups for conducting the breaching experiments using the WOZ technique.
Figure 3: Top: Wizard of Oz simulation.
3.4 Breaching Experiments
We have performed variations of seven breaching experiments, violating what we perceive as the norm5: 1. Robot was stuck and was trying to get out.
2. Robot was approaching people from the front, behind or from the side.
3. Robot was driving with abrupt and unpredictable behaviour.
4. Robot was standing still (no movement) in the middle of the pathway, blocking people’s natural walking direction.
5. Robot was using sound to get people’s attention.
6. Robot was blocked by an e-scooter.
7. Robot was ‘dressed up’ as a mobile defibrillator (AED).
3.5 Participants and Data Collection
A total of 338 encounters were recorded. Participants were recruited in situ as they incidentally and spontaneously passed and interacted with the robot. No prior information about the experiment was given to the InCoPs involved.
While one researcher was acting as the ‘wizard’, secretly controlling the robot remotely, the other researcher observed, took field notes, and conducted short in situ user interviews immediately after the robot interaction. The length of the interview varied from 3 to 15 minutes. During the data collection, special attention was paid to the process of membership categorisation. Interview questions were formulated beforehand but not rigorously followed6. Instead, the approach to interviewing was more of a contextual inquiry as proposed by [61], which allows the interviewer to pursue promising avenues of conversation. A contextual inquiry treats the interviewee as more of a partner, inviting the interviewee to ask questions. The promise is that this loose approach to interviewing allows the interviewer to “learn about the expectations and assumptions behind the user’s behaviours” [61, p. 3].
3.6 Method of Analysis
The method of analysis was based on an inductive approach to understanding the interaction, and an abductive approach to understanding how people ‘make sense’ of robots. Our aim has been to deconstruct and generalize accounts to make findings more accessible for designers. As such, it is in the method of analysis we conflict with the rigid
5 Videos of each breaching experiment can be found in Appendix 2.
6 Interview guide can be found in Appendix 5.
ethnomethodological way of analysing [62, 58]. Following the notion that pure ethnomethodology is somewhat
‘unconstructive’, we have adopted a more ‘constructive’ approach allowing ourselves to group and classify data throughout the analysis.
As it was not possible to do video recordings and transcribe every detail of interactions, we have not gone to the depth of transcribing and analysing every single interaction. Instead, we have used the conceptual framework as a lens to produce a story about the categories that InCoPs assign robots to, and the expectations that follow from these categories. Consequently, our approach to analysis can be seen as more lightweight, less detailed and influenced by our own interpretations. On the other hand, it allowed for the collection and analysis of more encounters. Inspired by Paulsen [63], our approach to collecting and analysing data was as follows:
8. Collecting data (written field notes) from interviews and observations attending to behaviour and interaction.
9. Building collections of explicit and implicit mentions of membership categories found in data.
10. Dividing encounters into groups that showed similar behaviour.
11. Analysing and interpreting data to identify categories, hidden assumptions, and expectations of the robot.
3.7 Ethical Considerations
In order to assess how InCoPs experience robots in their everyday lives, it was necessary to only include participants, who did not know about the research purpose and were in the process of performing everyday activities. The lack of prior consent poses an ethical challenge, which has shaped our experimental design. Firstly, permission was granted to use the area for conducting the experiments. Secondly, the robot was controlled remotely by a researcher using a Wizard of Oz simulation, which allowed for safe navigation and quick intervening in case of problems. Additionally, the researcher observing and taking field notes was always nearby and ready to intervene. The speed of the robot was set to a maximum of 6 kmph, which conforms with drafted regulations of autonomous guided robots in Denmark. Research ethics also constrained our choice of data collection methods. Video and audio recordings are usually preferred when conducting such studies, but due to the GDPR we only used written field notes from observations and interviews. Finally, all participants were debriefed, and the ‘wizard’ was revealed after each human-robot interaction.
4 RESULTS
In the following section, we will present the results from the study conducted over the three-week period in Copenhagen.
Initially, we will briefly run through the different types of InCoPs identified, as well as the membership categories which people used to make sense of the robot. In the second part of the results section, we will draw lines between the two, and present our results of how InCoPs interacted with and made sense of the robot as they encountered it. Finally, we will present our methodological findings.
4.1 Types of InCoPs
From the analysis of a total of 338 field encounters, the participants were grouped into six different types of InCoPs.
Each type could possibly be divided into additional subgroups, but these six groups represented the clearest distinction among the participants in terms of their behaviour and expectations. We further believe that they represent an appropriate balance between complexity and generalisation to inform our research agenda.
4.1.1 Children.
Participants aged 4 - 14 years walking through the public space. 37 of the 338 encounters (11%) were with children. A few children were on their own (mainly the older children), whereas most of them were present with a parent or caregiver.
4.1.2 Adults.
Participants aged 15 - 60 years walking through the public space. 186 encounters (55%) were with adults. They were either on their own or in a group.
4.1.3 Senior Citizens.
Participants aged 60+ years walking through the public space. 48 encounters (14%) were with senior citizens. A few were in groups of two, but the majority were on their own.
4.1.4 Caregivers.
Adults who were either parents looking after their child, or caregivers looking after a group of children. This group also included people walking their dogs. 27 encounters (8%) were with caregivers.
4.1.5 Cyclists.
30 cyclists (9%) participated in the field study. This group encountered the robot as they were either cycling in the pedestrian lane or cycling in the dedicated bike lane just next to the pedestrian lane where the robot was operating.
4.1.6 Dogs.
10 dogs (3%) participated in the field study. The dogs which interacted with the robot were all with their owner. Half of the dogs were on a lead, and half were off the lead. One dog was a trained guide dog.
4.2 Robot Membership Categories
From analysis of the verbal and non-verbal actions performed by InCoPs, we identified seven overall categories, which InCoPs typically assigned the robot to. People in some instances assigned the robot to multiple categories, and the categories often evolved or changed as the interaction unfolded.
4.2.1 Robot with an Unknown Purpose.
When people assigned the robot to this category, they acknowledged that what they saw was a robot, but they had no idea what it did, what it could do, and who it belonged to.
4.2.2 Pet.
When people assigned the robot to the ‘pet’ category, they predominantly saw the robot as a ‘living’ and ‘feeling’
creature with a social purpose.
4.2.3 Threat.
When the robot was seen as a ‘threat’, it was either perceived it as real threat (e.g. “I honestly thought it was a bomb”,
#252), as a potential future threat (e.g. “Someone could use that technology to do harm”, #18), or as not being an actual threat (e.g. “Look! Is that a bomb?” said jokingly to a friend, #10).
4.2.4 Entertainment.
When the robot was seen as a source of ‘entertainment’ or ‘fun’, people in some instances saw the element of entertainment as being the robot’ sole purpose (e.g. remote-controlled toy). However, more often the element of entertainment was seen as an addition to its actual purpose (e.g. delivering food). People were often aware that the sole purpose of the robot was not to entertain, but as the actual purpose was often irrelevant to them, they chose to see the technology as a source of entertainment, rather than a working robot (e.g. food delivery).
4.2.5 Working Robot.
The working robot category was assigned when participants believed the robot had a ‘practical job’, such as cutting the grass, last mile delivery or measuring the environment.
4.2.6 ’Dead’ Object.
When the robot did not move, it was generally considered a ‘dead’ object.
4.2.7 Experiment.
When the robot was categorised as an experiment, people saw it as an expensive prototype being tested by someone with connection to the university.
4.3 How InCoPs Experience Robots in Natural Environments
Table 1 brings an overview of the results of how the different InCoPs most commonly categorised the robot to make sense of it. The documentation of their expectations and membership categorisation during the interaction and post interaction interview will be presented in the following section.
Table 1: Membership categorisation based on type on InCoP.
Children Adults Senior Citizens Caregivers Cyclists Dogs
Unknown robot Unknown robot Unknown robot Pet Unknown robot Threat
Pet Working robot Working robot Threat ‘Dead’ object
Entertainment Pet Pet Entertainment
Threat Threat
Entertainment Entertainment
‘Dead’ object ‘Dead’ object Experiment
4.3.1 Children.
Children were quick at categorising the robot as either a ‘pet’, ‘entertainment’, or ‘robot with unknown purpose’, adopting behaviour that quickly made them engage in a more focused interaction.
Pet. Many children adopted behaviour and expectations that are similar to what can be seen in pets. One child asked their mom while pointing at the robot: "Where are it’s parents?" (#286), whereas another child was heard saying: “Don’t make it sad” (#11). These children showed a tendency to actively seek an interaction with the robot by moving closer, talking to it, and occasionally touching, tapping, or stroking it. This categorisation formed shared expectations of the robot as being an emotional, social, interactive and kind actor. As such, sounds, movements towards the children and reactions to their presence were expected and welcomed and led the children to engage in a more focused interaction.
Even when the robot did not comply with these expectations, for example by being inactive, children usually maintained their categorisation as a ‘pet’. As they had to move on, they would say things, such as “Bye bye” (#11) or “I hope to see you tomorrow” (#36).
Entertainment. Other accounts led us to suggest that several children categorised the robot as an object for
‘entertainment’. Children were heard saying things, such as “How much does it cost?” (#336), “I want one” (#254), and
“Where can you buy it?” (#1). They quickly engaged in a focused interaction trying to play with the robot by running next to it. In this scenario, children revealed expectations of the robot to be fun, playful, and something that they could own. Similar to children categorising the robot as pet, sounds were expected, but also abrupt robot behaviour and higher speeds. These children expected somewhat aggressive and unpredictable behaviour to conform with expectations of the robot as being playful.
Robot with Unknown Purpose. Some children showed behaviour that indicated that they had a hard time making sense of the robot. It seemed unclear to them what they could expect, which led us to interpret this categorisation as a
‘robot with unknown purpose’. Some showed curiosity, while others were reluctant to interact and seemed slightly worried, whereas other children tried to get more information by ‘testing’ it. One child kept pulling the antennas (#1), and another kept pushing the ‘stop’ button placed on top of the robot (#2).
